Can I run local LLMs on iPhone and Mac?

Yes. On Device AI supports local models on iPhone, iPad, Mac, and Vision Pro, with available models shaped by device capability.

What engines does On Device AI use?

On Device AI supports GGUF through llama.cpp and MLX models optimized for Apple Silicon.

Can I import custom GGUF models?

Yes. On Device AI supports custom GGUF imports from sources such as Hugging Face.

Run a Local LLM on iPhone, iPad and Mac

Local LLM settings and GGUF parameters on macOS

1. Choose an inference engine

On Device AI supports two local model formats. Available choices depend on the device and model you select.

GGUF via llama.cpp: A widely used format for compatible open-weight models.
MLX: A model format designed for Apple Silicon.

2. Pick a model your device can run

Model availability and performance depend on the model, quantization, and device memory. Start with a compatible option from the built-in catalog, then adjust your choice as your workload changes.

3. Custom Hugging Face GGUF Imports

You can import compatible custom GGUF models alongside the built-in catalog. Use the app's import flow and confirm that the model fits your device before relying on it for a workflow.

4. Keep local work local

Local model workflows process prompts and selected files on the device. No account is required for local workflows, and cloud providers are optional. See the Privacy Policy for details about optional connected services.

Download On Device AI Read Model Setup Guide →